What is the difference between 2-way and 3-way matching?

2-way match compares the supplier invoice against the purchase order, checking that price and quantity agree. 3-way match adds a third document, the goods receipt, so you confirm you were billed only for what you ordered and actually received. 3-way is stricter and catches more, which is why it is standard for physical goods.

What are matching tolerances?

Tolerances are the allowed difference before a line is treated as a mismatch, for example a small price rounding or a minor quantity variance. They let the system auto-approve invoices that agree closely enough while still flagging the ones with a real discrepancy, so a person only sees exceptions.

Will this replace our AP staff?

No. It changes what they do. Instead of matching every invoice line by hand, they handle the exceptions the system flags. The same team processes far more volume, and people spend their time on genuine discrepancies and supplier issues rather than routine matching.

How to automate invoice matching (2-way and 3-way)

Q: How does automated invoice matching work?

The system extracts the invoice fields, finds the matching PO and goods receipt, and compares price, quantity and totals within defined tolerances. If everything agrees inside the tolerance, the invoice posts through straight-through processing. If a line falls outside, a discrepancy workflow routes it to a person to resolve.

Automating invoice matching means the system checks each invoice against its purchase order (2-way) and its goods receipt (3-way), posts the invoices that agree within a set tolerance, and routes only the mismatches to a person. Instead of a clerk comparing three documents line by line for every invoice, one reviewer handles the exceptions. That single change is where the speed and the cost savings come from.

Here is how 2-way and 3-way matching work, what “automated” actually adds, and how to start without ripping out your accounts-payable process.

What is 2-way vs 3-way matching?

Both are ways to make sure you only pay for what you agreed to.

2-way match compares the supplier invoice against the purchase order. Do the price and quantity billed match what you ordered? If yes, the invoice is cleared to pay.
3-way match adds a third document: the goods receipt. Now you confirm three things agree: what you ordered (PO), what you received (receipt), and what you were billed (invoice). It catches the case where you were invoiced for goods that never arrived.

3-way is stricter and standard for physical goods, because it closes the gap between “ordered” and “received.” For services, a 2-way match against the PO or contract is often the practical version.

How does automated invoice matching work?

Done manually, matching means a person pulling up the PO and the receipt, and comparing them to the invoice field by field. Automated, the same logic runs as a pipeline:

Extract the invoice fields: vendor, invoice number, line items, quantities, prices, tax and total. (If your invoices arrive as PDFs, this is the data extraction step.)
Find the match: locate the corresponding PO and, for 3-way, the goods receipt.
Compare within tolerance: check price, quantity and totals against the PO and receipt, allowing for defined tolerances.
Decide: if everything agrees inside tolerance, the invoice posts straight through. If a line falls outside, a discrepancy workflow routes it to a person with the mismatch highlighted.

The system does not try to be clever about the exceptions. It is clear about what agrees and what doesn’t, and it puts a human only on the doesn’t.

What tolerances and exceptions actually do

Tolerances are the heart of it. A tolerance is the allowed difference before a line counts as a mismatch: a few cents of rounding, a small quantity variance, a freight charge within an expected band. Set them well and the system auto-approves the invoices that agree closely enough, while still flagging the ones with a real problem.

This is the same principle as confidence scoring in document automation, applied to AP: the system knows which invoices it can trust and escalates only the doubtful ones. We wrote about why that escalation step, not raw automation, is what makes accuracy safe in what 97% reconciliation accuracy really takes. Too tight a tolerance and everything becomes an exception; too loose and errors slip through. Tuning that band to your business is most of the work, and most of the value.

Where matching breaks, and what to do

Most invoices match cleanly. The value is in how the system handles the ones that do not, and the break cases are predictable:

Partial deliveries. The invoice covers goods that arrived in two shipments. The match has to net multiple receipts against one PO line, not expect a one-to-one.
Price and quantity variances. A small price change or short-ship should clear within tolerance; a large one should stop. This is exactly what the tolerance band is for.
Services with no goods receipt. A 3-way match has nothing to receive against, so these fall back to a 2-way match on the PO or contract.
Duplicate invoices. The same invoice submitted twice is a common source of overpayment. A duplicate check on vendor plus invoice number catches it before it posts.
Unit-of-measure mismatches. The PO is in cases, the invoice in units. Without a conversion rule, every line looks like a discrepancy.

Each of these is a rule you add once and reuse. The first weeks of a matching build are mostly about discovering which break cases your suppliers actually generate, then encoding them so they stop reaching a person.

What automated matching is worth

The numbers are stark once you stop matching agreeing invoices by hand. AP professionals process roughly 12 invoices per hour manually, versus about 30 with automation, more than double the throughput. On cost, manual AP processing runs $15–25 per invoice, while an automated environment handles the same invoice for under $5.

Those gains do not come from working faster. They come from not doing the work at all on the invoices that already agree, which is most of them. A person’s time shifts to the genuine discrepancies, supplier disputes and the judgment calls that actually need a human. We break down the full cost of doing this by hand in what manual invoice reconciliation actually costs.

How to start without a rebuild

You do not need to automate every supplier on day one:

Start with PO-backed, high-volume vendors. They match cleanly and give you the biggest straight-through rate immediately.
Set tolerances deliberately and watch the exception rate. If too much is flagging, your tolerances or your PO data need work, not the matching engine.
Extend to trickier suppliers (partial deliveries, services, no clean PO) once the core is running, adding rules as you go.

To size what automated matching could save you, our reconciliation ROI calculator turns your invoice volume and handling time into an estimate. And if you want us to look at your actual AP flow and tell you honestly where matching should be automated and where it shouldn’t, that is what the free 30-minute ROI diagnostic is for.

The teams that get the most out of it treat the exception queue as a signal, too. A supplier that keeps flagging is usually a data or contract problem worth fixing at the source, not just a matching miss, and clearing it upstream shrinks next month’s queue. Over time the exceptions tell you where your buying process itself is leaky.

The goal is not a system that never needs a person. It is one where a person only sees the invoices that genuinely disagree.